[dpdk-dev] [PATCH 3/4] net/hns3: fix VLAN tag inserted in wrong position in Tx

Wei Hu (Xavier) xavier.huwei at huawei.com
Sat Jul 4 12:09:50 CEST 2020


Based on hns3 network engine, in order to configure hardware VLAN insert
offload in Tx direction, PMD driver reads the VLAN tags from the
vlan_tci_outer and vlan_tci of the structure rte_mbuf, fills them into the
Tx Buffer Descriptor and sets the related offload flag for every packet.

Currently, there are two VLAN related problems in the 'tx_pkt_burst' ops
implementation function:
1) When setting the related offload flag, PMD driver inserts the VLAN tag
   into the position that close to L3 header. So, when upper application
   sends a packet with a VLAN tag in the data buffer, the VLAN offloaded
   by hardware will be added to the wrong position. It is supposed to add
   the VLAN tag from the rte_mbuf to the position close to the MAC header
   in the packet when using VLAN insertion.

   And when PF PVID is enabled by calling the API function named
   rte_eth_dev_set_vlan_pvid or VF PVID is enabled by hns3 PF kernel ether
   driver, the VLAN tag from the structure rte_mbuf to enable the VLAN
   insertion should be filled into the position that close to L3 header to
   avoid to be overwittern by the PVID which will always be inserted in the
   position that close to the MAC address.

2) When sending multiple segment packets, VLAN information is required to
   be filled into the first Tx Buffer descriptor. However, currently hns3
   PMD driver incorrectly placed it in the last Tx Buffer Descriptor. This
   results in VLAN insert offload failure when sending multiple segment
   packets.

This patch fixed them by filling the VLAN information into the position of
the Tx Buffer Descriptor.

Fixes: bba636698316 ("net/hns3: support Rx/Tx and related operations")
Cc: stable at dpdk.org

Signed-off-by: Chengchang Tang <tangchengchang at huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei at huawei.com>
Signed-off-by: Min Hu (Connor) <humin29 at huawei.com>
---
 drivers/net/hns3/hns3_rxtx.c | 102 +++++++++++++++++++++++++++----------------
 1 file changed, 65 insertions(+), 37 deletions(-)

diff --git a/drivers/net/hns3/hns3_rxtx.c b/drivers/net/hns3/hns3_rxtx.c
index 07640db..b0253d5 100644
--- a/drivers/net/hns3/hns3_rxtx.c
+++ b/drivers/net/hns3/hns3_rxtx.c
@@ -2007,51 +2007,58 @@ hns3_set_tso(struct hns3_desc *desc, uint64_t ol_flags,
 	desc->tx.mss = rte_cpu_to_le_16(rxm->tso_segsz);
 }
 
+static inline void
+hns3_fill_per_desc(struct hns3_desc *desc, struct rte_mbuf *rxm)
+{
+	desc->addr = rte_mbuf_data_iova(rxm);
+	desc->tx.send_size = rte_cpu_to_le_16(rte_pktmbuf_data_len(rxm));
+	desc->tx.tp_fe_sc_vld_ra_ri = rte_cpu_to_le_16(BIT(HNS3_TXD_VLD_B));
+}
+
 static void
-fill_desc(struct hns3_tx_queue *txq, uint16_t tx_desc_id, struct rte_mbuf *rxm,
-	  bool first, int offset)
+hns3_fill_first_desc(struct hns3_tx_queue *txq, struct hns3_desc *desc,
+		     struct rte_mbuf *rxm)
 {
-	struct hns3_desc *tx_ring = txq->tx_ring;
-	struct hns3_desc *desc = &tx_ring[tx_desc_id];
-	uint8_t frag_end = rxm->next == NULL ? 1 : 0;
 	uint64_t ol_flags = rxm->ol_flags;
-	uint16_t size = rxm->data_len;
-	uint16_t rrcfv = 0;
 	uint32_t hdr_len;
 	uint32_t paylen;
-	uint32_t tmp;
 
-	desc->addr = rte_mbuf_data_iova(rxm) + offset;
-	desc->tx.send_size = rte_cpu_to_le_16(size);
-	hns3_set_bit(rrcfv, HNS3_TXD_VLD_B, 1);
-
-	if (first) {
-		hdr_len = rxm->l2_len + rxm->l3_len + rxm->l4_len;
-		hdr_len += (ol_flags & PKT_TX_TUNNEL_MASK) ?
+	hdr_len = rxm->l2_len + rxm->l3_len + rxm->l4_len;
+	hdr_len += (ol_flags & PKT_TX_TUNNEL_MASK) ?
 			   rxm->outer_l2_len + rxm->outer_l3_len : 0;
-		paylen = rxm->pkt_len - hdr_len;
-		desc->tx.paylen = rte_cpu_to_le_32(paylen);
-		hns3_set_tso(desc, ol_flags, paylen, rxm);
-	}
-
-	hns3_set_bit(rrcfv, HNS3_TXD_FE_B, frag_end);
-	desc->tx.tp_fe_sc_vld_ra_ri = rte_cpu_to_le_16(rrcfv);
-
-	if (frag_end) {
-		if (ol_flags & (PKT_TX_VLAN_PKT | PKT_TX_QINQ_PKT)) {
-			tmp = rte_le_to_cpu_32(desc->tx.type_cs_vlan_tso_len);
-			hns3_set_bit(tmp, HNS3_TXD_VLAN_B, 1);
-			desc->tx.type_cs_vlan_tso_len = rte_cpu_to_le_32(tmp);
-			desc->tx.vlan_tag = rte_cpu_to_le_16(rxm->vlan_tci);
-		}
+	paylen = rxm->pkt_len - hdr_len;
+	desc->tx.paylen = rte_cpu_to_le_32(paylen);
+	hns3_set_tso(desc, ol_flags, paylen, rxm);
 
-		if (ol_flags & PKT_TX_QINQ_PKT) {
-			tmp = rte_le_to_cpu_32(desc->tx.ol_type_vlan_len_msec);
-			hns3_set_bit(tmp, HNS3_TXD_OVLAN_B, 1);
-			desc->tx.ol_type_vlan_len_msec = rte_cpu_to_le_32(tmp);
+	/*
+	 * Currently, hardware doesn't support more than two layers VLAN offload
+	 * in Tx direction based on hns3 network engine. So when the number of
+	 * VLANs in the packets represented by rxm plus the number of VLAN
+	 * offload by hardware such as PVID etc, exceeds two, the packets will
+	 * be discarded or the original VLAN of the packets will be overwitted
+	 * by hardware. When the PF PVID is enabled by calling the API function
+	 * named rte_eth_dev_set_vlan_pvid or the VF PVID is enabled by the hns3
+	 * PF kernel ether driver, the outer VLAN tag will always be the PVID.
+	 * To avoid the VLAN of Tx descriptor is overwritten by PVID, it should
+	 * be added to the position close to the IP header when PVID is enabled.
+	 */
+	if (!txq->pvid_state && ol_flags & (PKT_TX_VLAN_PKT |
+				PKT_TX_QINQ_PKT)) {
+		desc->tx.ol_type_vlan_len_msec |=
+				rte_cpu_to_le_32(BIT(HNS3_TXD_OVLAN_B));
+		if (ol_flags & PKT_TX_QINQ_PKT)
 			desc->tx.outer_vlan_tag =
-				rte_cpu_to_le_16(rxm->vlan_tci_outer);
-		}
+					rte_cpu_to_le_16(rxm->vlan_tci_outer);
+		else
+			desc->tx.outer_vlan_tag =
+					rte_cpu_to_le_16(rxm->vlan_tci);
+	}
+
+	if (ol_flags & PKT_TX_QINQ_PKT ||
+	    ((ol_flags & PKT_TX_VLAN_PKT) && txq->pvid_state)) {
+		desc->tx.type_cs_vlan_tso_len |=
+					rte_cpu_to_le_32(BIT(HNS3_TXD_VLAN_B));
+		desc->tx.vlan_tag = rte_cpu_to_le_16(rxm->vlan_tci);
 	}
 }
 
@@ -2628,8 +2635,10 @@ hns3_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 	struct rte_net_hdr_lens hdr_lens = {0};
 	struct hns3_tx_queue *txq = tx_queue;
 	struct hns3_entry *tx_bak_pkt;
+	struct hns3_desc *tx_ring;
 	struct rte_mbuf *tx_pkt;
 	struct rte_mbuf *m_seg;
+	struct hns3_desc *desc;
 	uint32_t nb_hold = 0;
 	uint16_t tx_next_use;
 	uint16_t tx_pkt_num;
@@ -2644,6 +2653,7 @@ hns3_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 	tx_next_use   = txq->next_to_use;
 	tx_bd_max     = txq->nb_tx_desc;
 	tx_pkt_num = nb_pkts;
+	tx_ring = txq->tx_ring;
 
 	/* send packets */
 	tx_bak_pkt = &txq->sw_ring[tx_next_use];
@@ -2688,8 +2698,22 @@ hns3_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 			goto end_of_tx;
 
 		i = 0;
+		desc = &tx_ring[tx_next_use];
+
+		/*
+		 * If the packet is divided into multiple Tx Buffer Descriptors,
+		 * only need to fill vlan, paylen and tso into the first Tx
+		 * Buffer Descriptor.
+		 */
+		hns3_fill_first_desc(txq, desc, m_seg);
+
 		do {
-			fill_desc(txq, tx_next_use, m_seg, (i == 0), 0);
+			desc = &tx_ring[tx_next_use];
+			/*
+			 * Fill valid bits, DMA address and data length for each
+			 * Tx Buffer Descriptor.
+			 */
+			hns3_fill_per_desc(desc, m_seg);
 			tx_bak_pkt->mbuf = m_seg;
 			m_seg = m_seg->next;
 			tx_next_use++;
@@ -2702,6 +2726,10 @@ hns3_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 			i++;
 		} while (m_seg != NULL);
 
+		/* Add end flag for the last Tx Buffer Descriptor */
+		desc->tx.tp_fe_sc_vld_ra_ri |=
+				 rte_cpu_to_le_16(BIT(HNS3_TXD_FE_B));
+
 		nb_hold += i;
 		txq->next_to_use = tx_next_use;
 		txq->tx_bd_ready -= i;
-- 
2.7.4



More information about the dev mailing list